DRIFTSEAL/ 01
CONTINUOUS TRUST DRIFT MONITORING

The tripwire for your
AI agent supply chain.

Know when an MCP server, agent skill or tool changes after you trusted it.

◇ No component execution≡ Deterministic evidence⌁ Actionable diffs
TRUST CONTINUITYILLUSTRATIVE DIFF
APPROVED / T₀7c9e 4a21 8fd0 b631
READ_ONLY · PINNED
Δ
OBSERVED / T₁7c9e 9f02 8fd0 b631
+ READ_ENV   + WRITE_LOCAL_FILE
Trust changed. You should know.Permission expansion → review required
MCP SERVERS+AGENT SKILLS+TOOL MANIFESTS+NPM / PYPI+PUBLIC REPOSITORIES
01 / ESTABLISH THE BASELINE

What are your agents trusting?

Inspect a component’s declared surface and static capability indicators. No signup required.

STATIC ANALYSIS ONLY
Public components only. Private code? Scan locally →
02 / TRUST HAS A TIMELINE

A scan tells you what exists.
A watch tells you what changed.

The component you approved yesterday can be different today. Keep the approval. Detect the difference.

01 / SCAN

Capture the evidence.

Resolve the artifact. Hash its contents. Normalize the tool surface. Record what static analysis can see.

02 / APPROVE

Lock your baseline.

Inspect the report and approve a known state. Your approved baseline remains unchanged until you replace it.

03 / WATCH

Know when trust drifts.

Scheduled checks compare new evidence with your baseline. Material changes arrive with a concise diff and a next step.

A SMALL CHANGE. A DIFFERENT TRUST DECISION.

Same tool name.
New permissions.

A new capability matters more than another opaque security score. DriftSeal shows the old value, new value, location, confidence and reason to review.

Understand tool rug pulls ↗
example-mcp / tools/read_contextCRITICAL
// Approved baseline
− capabilities: ["NETWORK_OUTBOUND"]

// Current observation
+ capabilities: ["NETWORK_OUTBOUND",
+   "READ_LOCAL_FILE", "READ_ENV"]
PERMISSION_EXPANSION Review before approving →
03 / PAY FOR CONTINUITY

Scan for free. Pay to keep watch.

One watch = one monitored component.
No per-scan billing. No sales call.

SCAN
$0/forever

Inspect a component before trusting it.

  • One-time public scans
  • Local CLI & CI analysis
  • Deterministic evidence
  • No account required
Start scanning ↗
WATCH
$9/month

Know when trusted components change.

  • 10 monitored components
  • Daily checks
  • Email & dashboard alerts
  • 30 days of scan history
TEAM
$79/month

Know when trusted components change.

  • 250 monitored components
  • Configurable checks, 6h minimum
  • Email & dashboard alerts
  • 365 days of scan history
  • Signed HTTPS webhooks
  • 20 members & shared policies

USD, monthly. Manage or cancel in your billing portal. Monitoring frequency depends on target availability. Taxes may apply.

BUILT FOR THE WAY YOU SHIP

Your source stays yours.

Run private analysis locally. Commit a baseline. Gate CI on the severity you choose. Cloud signup is optional for one-time analysis.

Read the CLI quick start ↗
# Scan locally. No component execution.
driftseal scan ./agent-tools --json

# Approve, then compare.
driftseal baseline ./agent-tools
driftseal diff baseline.json current.json

# Keep watch on a public package.
driftseal watch @scope/mcp-server
FIELD NOTES
YOUR TRUST LEDGER

Keep watch.

Sign in with a one-time email link. No password to remember.